RedPajama
Benchmarks
Task NameDataset NameSOTA ResultTrendResults
RedPajama LLaMA pretraining corpus (evaluation slice)
0.62Perplexity (bits/byte)
15
RedPajama
26.756Perplexity
13
RedPajama 100GB web content
94SA
8
RedPajama Safety Evals (test)
93.4Safety Score (Avg)
7
RedPajama Generation Quality Prefixes (test)
32.4Standard Prefix Count
4